The client is one of the world’s leading oilfield services providers. They have been energizing the world for about 90 years now. Starting from electricity to oil, they have been dominating the industry with their cutting-edge solutions for reservoir characterization, drilling, production, and processing. With a humble beginning in the 20th century, today they are operating in more than 120 countries.
An error in the oil industry could cost millions of dollars which makes precision a critical engineering aspect. Our client being the global leader in oilfield services was looking for someone who would build a scalable solution that can deliver drill bit recommendations with an accuracy that was never achieved before. Pluto7 being awarded as the Partner of the Year 2019 by Google for Data & Analytics brought their expertise and experience to solve this complex project to a confident conclusion. Our team has delivered many complex predictive analytics solutions for fortune 500 companies, therefore, the client looked at Pluto7 as the most reliable Google Cloud partner to help them generate accurate recommendations for the drill bit category and bit bomb numbers for any given site.
Oil extraction is more complex than it seems. Earth’s crust is not uniform, it is made up of different types of rocks and drilling is not always done vertically, sometimes it becomes important that angular trajectories are adopted. As a result, the extraction process is dynamic, pushing the team to use different drill bits – from the first impact to reaching the oil reserves. Each bit was categorized with bit number and bit category. These bits were differentiated on the basis of 80-90 features like the number of cutters in the bit. Given the amount of bit categories, the bit repository had a huge number of variants, making it challenging to recognize and remember each of them.
Pluto7’s team started by understanding the bit selection methodology that is used by field engineers. Additionally, the data from different sources was centralized and cleaned to train the model. Since the client was already on Google Cloud, the data was moved to BigQuery for running analysis. The goal of the project was to deliver accurate recommendations to the client with the set of required bits for a given time. The project was delivered with three approaches.
In the first approach the bit category was suggested on the basis of data fed to the prediction model. Bit clusters were made to narrow down the focus so that accurate bit numbers could be identified. Second approach was created to deliver results on a more granular level by recommending the bit numbers. Finally the third approach was built based on the historical activity, it gave suggestions on the basis of the most used drill bits in the past. Finally, a pipeline was built with these three models where the outcome of the preceding approach acted as input for the approach under observation.
The consistency and accuracy in the data was introduced after centralization. Three approaches were delivered to accurately recommend drill bits to site engineers, each approach giving a variation in the scope of recommendations. Visualizations and cluster representations of bit categories gave a better understanding of given formation data of any given site.